The Quantization Benefits of Residual-Free Transformers

• URL: http://arxiv.org/abs/2605.25880v1
• Date: Mon, 25 May 2026 14:06:02 GMT
• Title: The Quantization Benefits of Residual-Free Transformers
• Abstract summary: We show that residual connections can drive transformer activations away from Gaussianity during training.<n>In language tasks, while there is a small drop in full precision performance, these models retain near-Gaussian activations and exhibit significantly improved robustness to low-bit quantization.
• Score: 32.749359970205695
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Large-scale transformer training and deployment are increasingly constrained by the transfer of activations, gradients, and optimizer states across accelerators. Low-bit quantization offers a natural remedy, but transformer activations are often heavy-tailed and outlier-dominated, making simple quantization highly lossy. We show that this difficulty is not only a property of the quantizer, but also of the architecture. Specifically, residual connections can drive transformer activations away from Gaussianity during training. Using controlled comparisons between residual and residual-free transformers, we demonstrate that this effect leads to substantially higher quantization error and accuracy degradation at low precision in residual models. We explain the phenomenon through an excess kurtosis analysis, showing that residual mixing can amplify non-Gaussianity, whereas dense mixing in residual-free contracts non-Gaussianity. We then show that residual-free transformers can be made trainable using orthogonal initialization, spectral or second-order optimization, and depth-aware scaling of attention temperature. In language tasks, while there is a small drop in full precision performance, these models retain near-Gaussian activations and exhibit significantly improved robustness to low-bit quantization. Our results identify an accuracy--compressibility trade-off in transformer design and motivate architecture-level approaches to quantization-friendly foundation models.

Related papers

• Sharp Capacity Scaling of Spectral Optimizers in Learning Associative Memory [77.27772368491698]
  Spectrals such as Muon have recently shown strong empirical performance in large-scale language model training.<n>We study this question through the linear associative memory problem.<n>We show that the storage capacity of Muon significantly exceeds that of SGD.
  arXiv  Detail & Related papers  (2026-03-27T16:13:18Z)
• QuantSparse: Comprehensively Compressing Video Diffusion Transformer with Model Quantization and Attention Sparsification [67.15451442018258]
  Diffusion transformers exhibit remarkable video generation capability, yet their prohibitive computational and memory costs hinder practical deployment.<n>Model quantization and attention sparsification are two promising directions for compression, but each alone suffers severe performance degradation under aggressive compression.<n>We propose textbfQuantSparse, a unified framework that integrates model quantization with attention sparsification.
  arXiv  Detail & Related papers  (2025-09-28T06:49:44Z)
• Oscillations Make Neural Networks Robust to Quantization [0.16385815610837165]
  We show that oscillations in Quantization Aware Training (QAT) are undesirable artifacts caused by the Straight-Through Estimator (STE)<n>We propose a novel regularization method that induces oscillations to improve quantization.
  arXiv  Detail & Related papers  (2025-02-01T16:39:58Z)
• DopQ-ViT: Towards Distribution-Friendly and Outlier-Aware Post-Training Quantization for Vision Transformers [31.791935689364866]
  We propose DopQ-ViT, a Distribution-friendly and Outlier-aware Post-training Quantization method for Vision Transformers (ViTs)<n>First, DopQ-ViT introduces the Tan Quantizer (TanQ), which better preserves the power-law distribution of post-Softmax activations by focusing more on values near 1.<n>Second, DopQ-ViT presents the MAD-guided Optimal Scaling Factor (MOSF), which selects the optimal scaling factor without introducing additional calculations.
  arXiv  Detail & Related papers  (2024-08-06T16:40:04Z)
• Learning on Transformers is Provable Low-Rank and Sparse: A One-layer Analysis [63.66763657191476]
  We show that efficient numerical training and inference algorithms as low-rank computation have impressive performance for learning Transformer-based adaption. We analyze how magnitude-based models affect generalization while improving adaption. We conclude that proper magnitude-based has a slight on the testing performance.
  arXiv  Detail & Related papers  (2024-06-24T23:00:58Z)
• RepQuant: Towards Accurate Post-Training Quantization of Large Transformer Models via Scale Reparameterization [8.827794405944637]
  Post-training quantization (PTQ) is a promising solution for compressing large transformer models. Existing PTQ methods typically exhibit non-trivial performance loss. We propose RepQuant, a novel PTQ framework with quantization-inference decoupling paradigm.
  arXiv  Detail & Related papers  (2024-02-08T12:35:41Z)
• Teacher Intervention: Improving Convergence of Quantization Aware Training for Ultra-Low Precision Transformers [17.445202457319517]
  Quantization-aware training (QAT) is a promising method to lower the implementation cost and energy consumption. This work proposes a proactive knowledge distillation method called Teacher Intervention (TI) for fast converging QAT of ultra-low precision pre-trained Transformers.
  arXiv  Detail & Related papers  (2023-02-23T06:48:24Z)
• HEAT: Hardware-Efficient Automatic Tensor Decomposition for Transformer Compression [69.36555801766762]
  We propose a hardware-aware tensor decomposition framework, dubbed HEAT, that enables efficient exploration of the exponential space of possible decompositions. We experimentally show that our hardware-aware factorized BERT variants reduce the energy-delay product by 5.7x with less than 1.1% accuracy loss.
  arXiv  Detail & Related papers  (2022-11-30T05:31:45Z)
• NoisyQuant: Noisy Bias-Enhanced Post-Training Activation Quantization for Vision Transformers [53.85087932591237]
  NoisyQuant is a quantizer-agnostic enhancement for the post-training activation quantization performance of vision transformers. Building on the theoretical insight, NoisyQuant achieves the first success on actively altering the heavy-tailed activation distribution. NoisyQuant largely improves the post-training quantization performance of vision transformer with minimal computation overhead.
  arXiv  Detail & Related papers  (2022-11-29T10:02:09Z)
• Understanding and Overcoming the Challenges of Efficient Transformer Quantization [17.05322956052278]
  Transformer-based architectures have become the de-facto standard models for a wide range of Natural Language Processing tasks. However, their memory footprint and high latency are prohibitive for efficient deployment and inference on resource-limited devices. We show that transformers have unique quantization challenges -- namely, high dynamic activation ranges that are difficult to represent with a low bit fixed-point format.
  arXiv  Detail & Related papers  (2021-09-27T10:57:18Z)
• Post-Training Quantization for Vision Transformer [85.57953732941101]
  We present an effective post-training quantization algorithm for reducing the memory storage and computational costs of vision transformers. We can obtain an 81.29% top-1 accuracy using DeiT-B model on ImageNet dataset with about 8-bit quantization.
  arXiv  Detail & Related papers  (2021-06-27T06:27:22Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.